Research
Data and test sets about AI-written code, published with the method behind them and a download, so you can check our work and reuse it.
Everything here is licensed CC BY 4.0: use the data and charts anywhere, with a link to the page you took them from. Each page has a suggested citation.
- AI agent pull requests on GitHub, 2026
How many pull requests GitHub Copilot coding agent, Devin, Jules and Claude Code opened on public GitHub from January to September 2026, how many were merged, how fast, and how many with no review.
Method: Total counts from the public GitHub search API per agent, month and outcome, each agent identified by a bot author or attribution line its own docs describe. Every query is published.
- EU Cyber Resilience Act: timeline, SBOM rules and a checklist
The CRA's dates for software teams (in force 10 December 2024, reporting from 11 September 2026, full application 11 December 2027), the 24-hour reporting duty, the SBOM rules and a checklist.
Method: Every date and article reference checked against the text of Regulation (EU) 2024/2847 on EUR-Lex. A plain-language summary, not legal advice.
- Prompt injection test fixtures for coding agents
Twelve harmless fixtures that try to make a coding agent leak environment variables, skip tests or change CI, each with the safe behaviour and the control that catches it.
Method: Written by hand from the places injected text reaches a coding agent. A test set, not a benchmark: no agent scores are published.
- The AI dev-tools landscape 2026
71 AI developer tools in 10 vendor-neutral categories, from coding agents and AI code review to SBOM, secrets and quality gates, each linked to its official site. An embeddable map.
Method: Every official site opened and its own description checked against the category; alphabetical within each category, not a ranking. A script re-checks the links.
Writing about AI coding? The press page has our boilerplate and logos, and the free tools run in your browser.